pr4886: build-id enforcement

pr4886: build-id enforcement

[Frank Ch. Eigler]

Hi -

It seems that Wenji's pr4886 code is failing on fedora due to probable
elfutils and/or kernel build bugs.  Shall we disable these tests until
the tools/builds are fixed?  Even a mistaken warning message is
probably too much to punish fedora users with.  For now, how about
just sending out a printk and continuing, in case of a mismatch?

- FChE

Re: pr4886: build-id enforcement

[Elena Zannoni]

Frank Ch. Eigler wrote:
> Hi -
>
> It seems that Wenji's pr4886 code is failing on fedora due to probable
> elfutils and/or kernel build bugs.  Shall we disable these tests until
> the tools/builds are fixed?  Even a mistaken warning message is
> probably too much to punish fedora users with.  For now, how about
> just sending out a printk and continuing, in case of a mismatch?
>
> - FChE
>
>   

What is the ETA for a fix of the broken tools and builds? Where is the 
real problem?
Do you have more details?  Non-fedora users should find the build-id fix 
useful, even if
the fedora builds are broken.

elena

-- 
Elena Zannoni, Oracle
Senior Engineering Manager, Tools/Languages - Linux Engineering
Blog: http://blogs.oracle.com/ezannoni
Email: elena.zannoni@oracle.com

Re: pr4886: build-id enforcement

[Mark Wielaard]

Hi Elena,

On Mon, 2008-10-06 at 17:50 -0400, Elena Zannoni wrote:
> > It seems that Wenji's pr4886 code is failing on fedora due to probable
> > elfutils and/or kernel build bugs.  Shall we disable these tests until
> > the tools/builds are fixed?  Even a mistaken warning message is
> > probably too much to punish fedora users with.  For now, how about
> > just sending out a printk and continuing, in case of a mismatch?
>
> What is the ETA for a fix of the broken tools and builds? Where is the 
> real problem? Do you have more details?

I don't have much details. I did revert the patch locally though since
it does indeed seem to break things on fedora 9/rawhide (x86/x86_64) for
me. This is the bug to watch: "kernel build-id note corruption"
https://bugzilla.redhat.com/show_bug.cgi?id=465873

Cheers,

Mark

Re: pr4886: build-id enforcement

[Mark Wielaard]

Hi,

On Mon, 2008-10-06 at 17:18 -0400, Frank Ch. Eigler wrote:
> It seems that Wenji's pr4886 code is failing on fedora due to probable
> elfutils and/or kernel build bugs.  Shall we disable these tests until
> the tools/builds are fixed?  Even a mistaken warning message is
> probably too much to punish fedora users with.

At least the testsuite should be adapted to tolerate the new warnings.
Currently when the build-id is missing various tests see the "WARNING:
skip checking <module|exe-name>" and fail during make installcheck.
(And maybe the warning should include the word build-id to make it more
clear what is being skipped.)

Cheers,

Mark

Re: pr4886: build-id enforcement

[Roland McGrath]

The bug is in the Fedora kernel.spec magic that rejiggers build IDs with
debugedit.  It stopped doing it right for vmlinux, so that the vmlinux load
image embedded in /boot/vmlinuz-* (what actually gets booted) has a
different build ID than the file installed in /usr/lib/debug/.../vmlinux.
I'll update https://bugzilla.redhat.com/show_bug.cgi?id=465873 when the
problem is fixed.


Thanks,
Roland

Re: pr4886: build-id enforcement

[Wenji Huang]

Mark Wielaard wrote:
> Hi,
> 
> On Mon, 2008-10-06 at 17:18 -0400, Frank Ch. Eigler wrote:
>> It seems that Wenji's pr4886 code is failing on fedora due to probable
>> elfutils and/or kernel build bugs.  Shall we disable these tests until
>> the tools/builds are fixed?  Even a mistaken warning message is
>> probably too much to punish fedora users with.
> 
> At least the testsuite should be adapted to tolerate the new warnings.
> Currently when the build-id is missing various tests see the "WARNING:
> skip checking <module|exe-name>" and fail during make installcheck.
> (And maybe the warning should include the word build-id to make it more
> clear what is being skipped.)
> 
> Cheers,
> 
> Mark
> 
Sorry, the "skip checking" warning message is a little annoying. It 
should be moved backward or removed.

Frank,
thanks for simplifying it. But some cases are skipped, such as
no build-id in debuginfo, but exists in module/kernel.

Regards,
Wenji

Re: pr4886: build-id enforcement

[Frank Ch. Eigler]

Hi -

On Wed, Oct 08, 2008 at 09:45:19AM +0800, Wenji Huang wrote:

> [...] But some cases are skipped, such as no
> build-id in debuginfo, but exists in module/kernel.

IMO, these cases are out of the control of the user, and by nature
cannot contribute to build-id validation.  So we can lump them into
the other case where there is no build-id at all.

- FChE